From: Yen Chi Hsuan Date: Tue, 23 Feb 2016 04:17:41 +0000 (+0800) Subject: [nba] Support channels (#5362, #4167) X-Git-Url: http://git.oshgnacknak.de/?a=commitdiff_plain;h=3cfeb1624a28b60cec24dbc0d21565f4afd924b3;p=youtube-dl [nba] Support channels (#5362, #4167) --- diff --git a/youtube_dl/extractor/nba.py b/youtube_dl/extractor/nba.py index 87b38c4c7..3e2b3e599 100644 --- a/youtube_dl/extractor/nba.py +++ b/youtube_dl/extractor/nba.py @@ -1,11 +1,17 @@ from __future__ import unicode_literals +import functools import os.path import re from .common import InfoExtractor +from ..compat import ( + compat_urllib_parse, + compat_urlparse, +) from ..utils import ( int_or_none, + OnDemandPagedList, parse_duration, remove_start, xpath_text, @@ -60,16 +66,81 @@ class NBAIE(InfoExtractor): # m3u8 download 'skip_download': True, }, + }, { + 'url': 'http://www.nba.com/timberwolves/wiggins-shootaround#', + 'info_dict': { + 'id': 'timberwolves', + 'title': 'Shootaround Access - Dec. 12 | Andrew Wiggins', + }, + 'playlist_count': 30, + 'params': { + # Download the whole playlist takes too long time + 'playlist_items': '1-30', + }, + }, { + 'url': 'http://www.nba.com/timberwolves/wiggins-shootaround#', + 'info_dict': { + 'id': 'Wigginsmp4', + 'ext': 'mp4', + 'title': 'Shootaround Access - Dec. 12 | Andrew Wiggins', + 'description': 'Wolves rookie Andrew Wiggins addresses the media after Friday\'s shootaround.', + 'upload_date': '20141212', + 'timestamp': 1418418600, + }, + 'params': { + 'noplaylist': True, + # m3u8 download + 'skip_download': True, + }, }] + _PAGE_SIZE = 30 + + def _fetch_page(self, team, video_id, page): + search_url = 'http://searchapp2.nba.com/nba-search/query.jsp?' + compat_urllib_parse.urlencode({ + 'type': 'teamvideo', + 'start': page * self._PAGE_SIZE + 1, + 'npp': (page + 1) * self._PAGE_SIZE + 1, + 'sort': 'recent', + 'output': 'json', + 'site': team, + }) + results = self._download_json( + search_url, video_id, note='Download page %d of playlist data' % page)['results'][0] + for item in results: + yield self.url_result(compat_urlparse.urljoin('http://www.nba.com/', item['url'])) + + def _extract_playlist(self, orig_path, video_id, webpage): + team = orig_path.split('/')[0] + + if self._downloader.params.get('noplaylist'): + self.to_screen('Downloading just video because of --no-playlist') + video_path = self._search_regex( + r'nbaVideoCore\.firstVideo\s*=\s*\'([^\']+)\';', webpage, 'video path') + video_url = 'http://www.nba.com/%s/video/%s' % (team, video_path) + return self.url_result(video_url) + + self.to_screen('Downloading playlist - add --no-playlist to just download video') + playlist_title = self._og_search_title(webpage, fatal=False) + entries = OnDemandPagedList( + functools.partial(self._fetch_page, team, video_id), + self._PAGE_SIZE, use_cache=True) + + return self.playlist_result(entries, team, playlist_title) + def _real_extract(self, url): path, video_id = re.match(self._VALID_URL, url).groups() + orig_path = path if path.startswith('nba/'): path = path[3:] if 'video/' not in path: webpage = self._download_webpage(url, video_id) path = remove_start(self._search_regex(r'data-videoid="([^"]+)"', webpage, 'video id'), '/') + + if path == '{{id}}': + return self._extract_playlist(orig_path, video_id, webpage) + # See prepareContentId() of pkgCvp.js if path.startswith('video/teams'): path = 'video/channels/proxy/' + path[6:]